X-Original-To: pgsql-performance-postgresql.org@localhost.postgresql.org
Received: from localhost (av.hub.org [200.46.204.144])
	by svr1.postgresql.org (Postfix) with ESMTP id 70494D9ED2;
	Mon,  3 Oct 2005 18:32:38 -0300 (ADT)
Received: from svr1.postgresql.org ([200.46.204.71])
	by localhost (av.hub.org [200.46.204.144]) (amavisd-new, port 10024)
	with ESMTP id 42938-04; Mon,  3 Oct 2005 21:32:30 +0000 (GMT)
Received: from gghcwest.com (adsl-71-128-90-172.dsl.pltn13.pacbell.net
	[71.128.90.172])
	by svr1.postgresql.org (Postfix) with ESMTP id 96D45D9ED1;
	Mon,  3 Oct 2005 18:32:30 -0300 (ADT)
Received: from toonses.gghcwest.com (toonses.gghcwest.com [192.168.168.115])
	by gghcwest.com (8.12.10/8.12.9) with ESMTP id j93LWOmd013894;
	Mon, 3 Oct 2005 14:32:25 -0700
Received: from jwb by toonses.gghcwest.com with local (Exim 4.52)
	id 1EMXv8-0007Zr-Rm; Mon, 03 Oct 2005 14:32:26 -0700
Subject: Re: [HACKERS] A Better External Sort?
From: "Jeffrey W. Baker" <jwbaker@acm.org>
To: Josh Berkus <josh@agliodbs.com>
Cc: pgsql-hackers@postgresql.org, Michael Stone <mstone+postgres@mathom.us>,
	pgsql-performance@postgresql.org
In-Reply-To: <200510031416.15620.josh@agliodbs.com>
References: 
 <31833713.1128111650174.JavaMail.root@elwamui-polski.atl.sa.earthlink.net>
	<200510031334.02030.josh@agliodbs.com>
	<1128372151.28509.1.camel@toonses.gghcwest.com>
	<200510031416.15620.josh@agliodbs.com>
Content-Type: text/plain
Content-Transfer-Encoding: 7bit
Date: Mon, 03 Oct 2005 14:32:26 -0700
Message-Id: <1128375146.29080.9.camel@toonses.gghcwest.com>
Mime-Version: 1.0
X-Mailer: Evolution 2.4.0 
X-Virus-Scanned: by amavisd-new at hub.org
X-Spam-Status: No, hits=-0.361 required=5 tests=[AWL=-0.411,
	FORGED_RCVD_HELO=0.05]
X-Spam-Level: 
X-Archive-Number: 200510/30
X-Sequence-Number: 14798

On Mon, 2005-10-03 at 14:16 -0700, Josh Berkus wrote:
> Jeff,
> 
> > > Nope, LOTS of testing, at OSDL, GreenPlum and Sun.   For comparison, A
> > > Big-Name Proprietary Database doesn't get much more than that either.
> >
> > I find this claim very suspicious.  I get single-threaded reads in
> > excess of 1GB/sec with XFS and > 250MB/sec with ext3.
> 
> Database reads?  Or raw FS reads?  It's not the same thing.

Just reading files off the filesystem.  These are input rates I get with
a specialized sort implementation.  1GB/sec is not even especially
wonderful, I can get that on two controllers with 24-disk stripe set.

I guess database reads are different, but I remain unconvinced that they
are *fundamentally* different.  After all, a tab-delimited file (my sort
workload) is a kind of database.

> Also, we're talking *write speed* here, not read speed.

Ok, I did not realize.  Still you should see 250-300MB/sec
single-threaded sequential output on ext3, assuming the storage can
provide that rate.

> I also find *your* claim suspicious, since there's no way XFS is 300% faster 
> than ext3 for the *general* case.

On a single disk you wouldn't notice, but XFS scales much better when
you throw disks at it.  I get a 50MB/sec boost from the 24th disk,
whereas ext3 stops scaling after 16 disks.  For writes both XFS and ext3
top out around 8 disks, but in this case XFS tops out at 500MB/sec while
ext3 can't break 350MB/sec.

I'm hopeful that in the future the work being done at ClusterFS will
make ext3 on-par with XFS.

-jwb